Free tiers of popular artificial intelligence assistants charge users through data collection, advertisements, or stripped-down model performance rather than upfront subscription fees. A comparative test published by the German technology publication Notebookcheck examines the exact costs of using unpaid AI models from major developers, revealing that default settings frequently enroll users in data harvesting.
Data Collection Defaults on ChatGPT and Gemini
Users opening a free ChatGPT or Gemini account automatically consent to data harvesting unless they actively modify their privacy settings. OpenAI applies data collection by default to train its models on user interactions. Similarly, Google activates its “Keep Activity” setting by default for users over 18, allowing human reviewers to read a portion of conversations and store them for up to three years.
Google states in its privacy center that Gemini conversations do not serve targeted advertisements, yet the platform retains standard user activity for up to 72 hours even after users disable the activity log.
Advertising Models and Feature Restrictions
Monetization strategies vary significantly across the four major platforms evaluated in the study. OpenAI introduced advertisements to free tiers in Europe, where non-personalized notices appear alongside standard text generation. Microsoft adopts an aggressive monetization approach for Copilot, incorporating targeted advertisements derived from conversational history and saved memories unless users manually disable ad personalization in the settings dashboard.
Model capability restrictions are the primary limitation for unpaid users of the software. OpenAI limits free ChatGPT queries to a 27,000-token context window—roughly 12 pages of text—which stands at half the capacity offered by paid subscription tiers. Anthropic restricts its flagship Opus model to paying subscribers while offering access to Sonnet and Haiku variants on the free tier. Meanwhile, Google provides access to Gemini 3.6 Flash alongside limited features for Gemini 3.1 Pro on its zero-cost tier.
Users Adjust Settings or Run Local Models for Privacy
Adjusting account settings remains the primary method for individuals to protect sensitive data without paying for commercial subscriptions. Users can deactivate data training toggles within the privacy panels of ChatGPT, Claude, and Copilot. Microsoft retains Copilot conversations for 18 months by default unless adjusted, while Anthropic maintains ambiguity regarding its default data retention policies on free tiers.
Running open-source large language models locally on personal hardware provides a complete alternative to cloud-based assistants. This method requires a computer with sufficient processing power and an acceptance of lower performance compared to commercial offerings from OpenAI, Google, Anthropic, or Microsoft.
Keep reading